Re: [PATCH v2] rust: irq: add &Device<Bound> argument to irq callbacks

[Boqun Feng <boqun.feng@gmail.com>]

On Mon, Aug 11, 2025 at 04:25:50PM +0200, Alice Ryhl wrote:
> On Mon, Aug 11, 2025 at 4:24 PM Boqun Feng <boqun.feng@gmail.com> wrote:
> >
> > On Mon, Aug 11, 2025 at 04:05:31PM +0200, Alice Ryhl wrote:
> > > On Mon, Aug 11, 2025 at 3:56 PM Boqun Feng <boqun.feng@gmail.com> wrote:
> > > >
> > > > On Mon, Aug 11, 2025 at 12:33:51PM +0000, Alice Ryhl wrote:
> > > > [...]
> > > > > @@ -207,8 +207,8 @@ pub fn new<'a>(
> > > > >              inner <- Devres::new(
> > > > >                  request.dev,
> > > > >                  try_pin_init!(RegistrationInner {
> > > > > -                    // SAFETY: `this` is a valid pointer to the `Registration` instance
> > > > > -                    cookie: unsafe { &raw mut (*this.as_ptr()).handler }.cast(),
> > > > > +                    // INVARIANT: `this` is a valid pointer to the `Registration` instance
> > > > > +                    cookie: this.as_ptr().cast::<c_void>(),
> > > >
> > > > At this moment the `Regstration` is not fully initialized...
> > > >
> > > > >                      irq: {
> > > > >                          // SAFETY:
> > > > >                          // - The callbacks are valid for use with request_irq.
> > > > > @@ -221,7 +221,7 @@ pub fn new<'a>(
> > > > >                                  Some(handle_irq_callback::<T>),
> > > > >                                  flags.into_inner(),
> > > > >                                  name.as_char_ptr(),
> > > > > -                                (&raw mut (*this.as_ptr()).handler).cast(),
> > > > > +                                this.as_ptr().cast::<c_void>(),
> > > > >                              )
> > > >
> > > > ... and interrupt can happen right after request_irq() ...
> > > >
> > > > >                          })?;
> > > > >                          request.irq
> > > > > @@ -258,9 +258,13 @@ pub fn synchronize(&self, dev: &Device<Bound>) -> Result {
> > > > >  ///
> > > > >  /// This function should be only used as the callback in `request_irq`.
> > > > >  unsafe extern "C" fn handle_irq_callback<T: Handler>(_irq: i32, ptr: *mut c_void) -> c_uint {
> > > > > -    // SAFETY: `ptr` is a pointer to T set in `Registration::new`
> > > > > -    let handler = unsafe { &*(ptr as *const T) };
> > > > > -    T::handle(handler) as c_uint
> > > > > +    // SAFETY: `ptr` is a pointer to `Registration<T>` set in `Registration::new`
> > > > > +    let registration = unsafe { &*(ptr as *const Registration<T>) };
> > > >
> > > > ... hence it's not correct to construct a reference to `Registration`
> > > > here, but yes, both `handler` and the `device` part of `inner` has been
> > > > properly initialized. So
> > > >
> > > >         let registration = ptr.cast::<Registration<T>>();
> > > >
> > > >         // SAFETY: The `data` part of `Devres` is `Opaque` and here we
> > > >         // only access `.device()`, which has been properly initialized
> > > >         // before `request_irq()`.
> > > >         let device = unsafe { (*registration).inner.device() };
> > > >
> > > >         // SAFETY: The irq callback is removed before the device is
> > > >         // unbound, so the fact that the irq callback is running implies
> > > >         // that the device has not yet been unbound.
> > > >         let device = unsafe { device.as_bound() };
> > > >
> > > >         // SAFETY: `.handler` has been properly initialized before
> > > >         // `request_irq()`.
> > > >         T::handle(unsafe { &(*registration).handler }, device) as c_uint
> > > >
> > > > Thoughts? Similar for the threaded one.
> > >
> > > This code is no different. It creates a reference to `inner` before
> > > the `irq` field is written. Of course, it's also no different in that
> > > since data of a `Devres` is in `Opaque`, this is not actually UB.
> > >
> >
> > Well, I think we need at least mentioning that it's safe because we
> > don't access .inner.inner here, but..
> >
> > > What I can offer you is to use the closure form of pin-init to call
> > > request_irq after initialization has fully completed.
> > >
> >
> > .. now you mention this, I think we can just move the `request_irq()`
> > to the initializer of `_pin`:
> >
> > ------>8
> > diff --git a/rust/kernel/irq/request.rs b/rust/kernel/irq/request.rs
> > index ae5d967fb9d6..3343964fc1ab 100644
> > --- a/rust/kernel/irq/request.rs
> > +++ b/rust/kernel/irq/request.rs
> > @@ -209,26 +209,26 @@ pub fn new<'a>(
> >                  try_pin_init!(RegistrationInner {
> >                      // INVARIANT: `this` is a valid pointer to the `Registration` instance
> >                      cookie: this.as_ptr().cast::<c_void>(),
> > -                    irq: {
> > -                        // SAFETY:
> > -                        // - The callbacks are valid for use with request_irq.
> > -                        // - If this succeeds, the slot is guaranteed to be valid until the
> > -                        //   destructor of Self runs, which will deregister the callbacks
> > -                        //   before the memory location becomes invalid.
> > -                        to_result(unsafe {
> > -                            bindings::request_irq(
> > -                                request.irq,
> > -                                Some(handle_irq_callback::<T>),
> > -                                flags.into_inner(),
> > -                                name.as_char_ptr(),
> > -                                this.as_ptr().cast::<c_void>(),
> > -                            )
> > -                        })?;
> > -                        request.irq
> > -                    }
> > +                    irq: request.irq
> >                  })
> >              ),
> > -            _pin: PhantomPinned,
> > +            _pin: {
> > +                // SAFETY:
> > +                // - The callbacks are valid for use with request_irq.
> > +                // - If this succeeds, the slot is guaranteed to be valid until the
> > +                //   destructor of Self runs, which will deregister the callbacks
> > +                //   before the memory location becomes invalid.
> > +                to_result(unsafe {
> > +                    bindings::request_irq(
> > +                        request.irq,
> > +                        Some(handle_irq_callback::<T>),
> > +                        flags.into_inner(),
> > +                        name.as_char_ptr(),
> > +                        this.as_ptr().cast::<c_void>(),
> > +                    )
> > +                })?;
> > +                PhantomPinned
> > +            },
> >          })
> >      }
> >
> >
> > Thoughts?
> 
> That calls free_irq if request_irq fails, which is illegal.
> 

Ah, right. I was missing that. Then back to the "we have to mention that
we are not accessing the data of Devres" suggestion, which I think is
more appropriate for this case.

Regards,
Boqun

> Alice